Address correspondence and reprint requests to A. M. Goate, Department of Psychiatry, Washington University School of Medicine, 660 S. Euclid Ave. B8134, St. Louis, MO 63110, USA. E-mail: firstname.lastname@example.org
Alzheimer's disease (AD) is the most common form of dementia and is characterized pathologically by the accumulation of β-amyloid (Aβ) plaques and neurofibrillary tangles in the brain. Genetic studies of AD first highlighted the importance of the presenilins (PS). Subsequent functional studies have demonstrated that PS form the catalytic subunit of the γ-secretase complex that produces the Aβ peptide, confirming the central role of PS in AD biology. Here, we review the studies that have characterized PS function in the γ-secretase complex in Caenorhabditis elegans, mice and in in vitro cell culture systems, including studies of PS structure, PS interactions with substrates and other γ-secretase complex members, and the evidence supporting the hypothesis that PS are aspartyl proteases that are active in intramembranous proteolysis. A thorough knowledge of the mechanism of PS cleavage in the context of the γ-secretase complex will further our understanding of the molecular mechanisms that cause AD, and may allow the development of therapeutics that can alter Aβ production and modify the risk for AD.
Genetic studies in families that have a history of AD (FAD) have identified four genes that contribute to AD: the ε4 allele of apolipoprotein E is a risk factor while causative mutations have been demonstrated in the genes encoding β- APP, presenilin-1 and -2 (PS1, PS2). Most mutations in APP and PS increase the ratio of a 42-residue form of Aβ (Aβ42) versus 40-residue Aβ (Aβ40) (Aβ42 : Aβ40), thus defining a common AD phenotype caused by APP, PS1 and PS2 mutations (Scheuner et al. 1996). The vast majority of these mutations have been reported in the PS genes (Cruts and Rademakers 2004). As a result, much research has focused on understanding the role of PS in APP metabolism and the molecular mechanisms that cause AD, with the goal of developing vaccines or medications that can modify APP metabolism and reduce the prevalence of AD.
Cell biology of presenilin
The first evidence that PS homologues exist in other species came from work in Caenorhabditis elegans (C. elegans). A screen for suppressors of a lin-12 (Notch) gain-of-function mutation identified sel-12, which is 48% identical to human PS1 (hPS1) at the protein level (Levitan and Greenwald 1995). hPS1 expressed in a C. elegans sel-12 mutant was able to rescue the Notch mutant phenotype, confirming that sel-12 is a PS (Levitan et al. 1996). C. elegans has a second PS, HOP-1, that is 33% identical to hPS1 (Li and Greenwald 1997) and can rescue the Notch mutant phenotype of sel-12. PS homologues have now been identified in other multicellular species ranging from mammals to frogs, flies, worms, fish and plants (Fig 2).
Semi-quantitative reverse-transcribed RNA analyses (RT-PCR) of RNA levels have demonstrated that PS1 and PS2 are both expressed throughout most adult human tissues and brain regions (Rogaev et al. 1995; Sherrington et al. 1995; Lee et al. 1996; Benkovic et al. 1997; Berezovska et al. 1997, 1998; Lah et al. 1997). Immunoblot and immunohistochemistry experiments have confirmed the distribution of PS1 and PS2 proteins throughout most peripheral tissues and all brain regions (Moussaoui et al. 1996; Blanchard et al. 1997; Lah et al. 1997). RT-PCR was also used to study mRNA levels in mouse tissues during development. PS1 is expressed earlier than PS2 during mouse embryonic development. PS1 mRNA levels are twice those of PS2 in E8.5 and E10.5 whole-mouse embryos, but by day E12.5, PS2 mRNA levels rise to match those of PS1. Comparable to expression levels in human tissues, PS1 mRNA accumulates to a level three times that of PS2 in the cortex of newborn mice, while adult mice have similar levels of PS1 and PS2 mRNA in the cortex. These data suggest that PS1 and PS2 transcript expression may be regulated differently during development and, importantly, during brain maturation (Lee et al. 1996).
Hydropathy plots identify 10 hydrophobic regions (HR), predicting that PS contain seven to 10 transmembrane domains (TM) and are integral membrane proteins. Experimental evidence for PS orientation in the membrane and number of membrane-spanning domains has been collected through two primary approaches: staining with antibodies, and the placement of tags at various locations through the PS molecule. These studies have produced varying structure predictions. Some studies predict that PS cross the membrane seven times. Although two reports place the N-terminus and large loop including HR7 in the lumen extracellularly while the C-terminus is localized intracellularly (Dewji and Singer 1997; Dewji et al. 2004), a third study showed that the N-terminus and large loop are expressed in the cytoplasm while the C-terminus is exposed extracellularly (Nakai et al. 1999). Other experiments predict that PS have an even number of transmembrane domains (six or eight) and that the N-terminus, large loop region and C-terminus all reside in the cytoplasm (Doan et al. 1996; Li and Greenwald 1996, 1997, 1998; De Strooper et al. 1997; Lehmann et al. 1997).
Studies of PS topology have thus failed to provide a clear picture of PS orientation, due to the numerous discrepancies between experiments. Immunohistochemistry with PS-specific antibodies in permeabilized or non-permeabilized cells provides different results in experiments by different groups, probably due to differences in antibody specificity. PS constructs that have been engineered to express glycosylation sites, antigens or reporter sequences at various locations throughout the PS molecules are highly artificial constructs. The addition of tags may itself influence insertion into the membrane, producing PS molecules with an abnormal conformation. For example, the HR constructs used in the Lehmann et al. (1997) experiments are not processed efficiently (A. L. Brunkan and A. M. Goate, unpublished data), and it has been shown by numerous groups (Thinakaran et al. 1996; Tomita et al. 1999b; Shirotani et al. 2000; Brunkan et al. 2005) that the addition of a tag at the C-terminus of PS results in reduced PS processing (for a discussion of endoproteolytic processing of PS see Presenilinase and post-translational processing of PS, below).
An eight TM structure is currently assumed based on the experiments in C. elegans reported by Li and Greenwald (1996, 1997, 1998) (Fig 3). The residues that comprise each HR have also not been defined. TMs in this schematic (Fig 3) were determined assuming a minimum requirement for 18 residues in a TM, using known TM insertion start and stop signals, evaluating the predictions made by four programs [Engelmen-stietz (uses hydrophobicity) by toppred, TMHMM, SOSUI and TMAP by Biology Workbench] and comparing these with previously proposed domain definitions (Crook 2000). HR7 and HR10 are depicted as domains that may form membrane-associated structures despite their inability to cross the membrane. Refinement of this predicted structure awaits further research.
Presenilinase and post-translational processing of PS
PS1 and PS2 are both phosphorylated on serine residues, but the significance of this modification is unknown (Walter et al. 1996, 1997; De Strooper et al. 1997; Seeger et al. 1997). The most important and well studied post-translational modification of PS1 and PS2 is proteolytic processing by an enzyme activity called presenilinase to generate an N-terminal fragment (NTF) and C-terminal fragment (CTF). When overexpressed in tissue culture systems, an approximately 43 kDa full-length (FL) PS1 molecule was identified along with an approximately 27 kDa NTF and an approximately 17 kDa CTF (Thinakaran et al. 1996). Endogenous PS1 in tissue culture cells and mouse, monkey and human brains was found predominantly in the NTF/CTF forms, and little if any FL-PS1 could be detected (Thinakaran et al. 1996). Similar fragments were observed for PS2, which is expressed as an approximately 54 kDa FL-PS2, an approximately 34 kDa NTF and an approximately 20 kDa CTF (Kim et al. 1997b; Podlisny et al. 1997). The PS1 and PS2 CTFs are also subject to caspase-mediated cleavage to produce an approximately 10–15 kDa fragment (Kim et al. 1997a; Loetscher et al. 1997; Brockhaus et al. 1998; Grunberg et al. 1998b). It has since been shown that the cleaved forms of PS are the primary species expressed endogenously in all systems analyzed [e.g. C. elegans (Baumeister et al. 1997) and zebrafish (Leimer et al. 1999)]. Several FAD mutations that result in deletion of exon nine (ΔE9 mutations; residues 290–319) have been identified that remove a portion of a large cytosolic region that is important for PS processing, resulting in a FL-PS1 molecule that is not processed to NTF and CTF fragments (Perez-Tur et al. 1995; Prihar et al. 1999). The deletion of exon nine is accompanied by a change in codon 290 from serine to cysteine (Perez-Tur et al. 1995; Prihar et al. 1999), and in vitro experiments have linked the AD phenotype in these families to the S290C point mutation rather than the exon nine deletion (ΔE9) (Steiner et al. 1999a). These data suggest that a FL-PS1 molecule (ΔE9) may not require endoproteolysis to NTF and CTF for its function in producing an AD phenotype.
Overexpression of hPS1 in transgenic mice demonstrated that the levels of the NTF/CTF are saturable, that NTF/CTF are always present in a 1 : 1 stoichiometry, and that overexpression does not increase the amount of NTF/CTF protein expression. Rather, in these mice (and other overexpression systems) the FL-PS1 accumulates, and high molecular weight aggregates are also visible. Furthermore, transgenic expression of hPS1 in mice causes a decrease in the amount of endogenous NTF/CTF, suggesting that the high level of expression of transgenic hPS1 causes a ‘replacement’ phenomenon in which the endogenous mouse PS1 is down-regulated (Thinakaran et al. 1996). Replacement has also been observed in tissue culture transfection experiments (e.g. De Strooper et al. 1997) and the extent of replacement varies inversely with the level of exogenous transgene expression (Thinakaran et al. 1997; Saura et al. 1999), suggesting that transgene and endogenous PS compete for a limiting cellular factor(s) that regulates PS expression levels (Thinakaran et al. 1997). (For a discussion of possible limiting factors, see γ-secretase is a multiprotein complex, below).
Pulse-chase studies have demonstrated that FL-PS1 has a half-life of approximately 1–2 h, while the NTF/CTF have a T1/2 of over 24 h (Podlisny et al. 1997; Ratovitski et al. 1997). Studies with PS2 indicate that the NTF/CTF levels are also stable for over 24 h (Kim et al. 1997b; Steiner et al. 1998). The ΔE9 mutant that does not undergo presenilinase cleavage is stable for over 40 h, indicating that endoproteolysis of PS is not essential for proper maturation, and that the FL-PS1 molecule may be stabilized and then endoproteolytically cleaved to form the NTF and CTF (Ratovitski et al. 1997). FL-PS2 is poly-ubiquitinated in a pre-Golgi compartment, and FL-PS1 and PS2 are degraded by the proteasomal pathway. However, proteasome inhibitors do not block the production of PS2 NTF/CTF or PS1 NTF/CTF, indicating that the ubiquitin-proteasome is not responsible for the presenilinase activity that cleaves FL-PS (Kim et al. 1997b; Steiner et al. 1998). The caspase-induced PS1 CTF is stabilized by cysteine protease inhibitors, demonstrating that this PS fragment is primarily removed by proteases (Steiner et al. 1998).
Separation of cell membrane preparations on iodixanol gradients demonstrated that FL-PS is present in the ER while the NTF and CTF are present primarily in the Golgi (Xia et al. 1998; Zhang et al. 1998a; Annaert et al. 1999). Furthermore, the unprocessed ΔE9 PS molecule is present in both the ER and Golgi fractions. The presence of FAD mutations (PS1/M146L, PS1/C410Y, PS2/N141I) does not affect FL-PS1 or fragment subcellular localization. These data suggest that presenilinase processing occurs in the ER and that transport to the Golgi may be important for PS stabilization (Zhang et al. 1998a). However, treatment with inhibitors that block presenilinase activity and cause the accumulation of endogenous FL-PS1 results in sorting of FL-PS1 to the Golgi as well as the ER, mirroring the distribution of NTF in sucrose density gradients. The ER accumulation in overexpression studies may be due to the restricted expression of a limiting stoichiometric factor that is necessary for PS to leave the ER. Apparently, presenilinase cleavage is not required for trafficking of PS through the secretory pathway (Beher et al. 2001).
The ΔE9 FAD variant does not undergo endoproteolysis, pinpointing the site of proteolysis within amino acids 290–320 of PS1 (Thinakaran et al. 1996). Radiosequencing identified the major cleavage site of PS2 between residues M298 and V299. The CTF produced from presenilinase cleavage of PS2 may undergo secondary cleavages at its N-terminus to generate a CTF beginning with residue L307 or, alternatively, a minor cleavage site may exist between residues K306 and L307 (Shirotani et al. 1997; Jacobsen et al. 1999). The double mutation M298D/V299A resulted in undetectable levels of PS2 NTF in stable N2a cells (Jacobsen et al. 1999), and deletion of residues 297–316 eliminated endoproteolysis in SH-SY5Y cells (Shirotani et al. 2000). Introduction of an FAD mutation in a PS2 mutant lacking residues 296–301 and containing a K306G mutation did not show the expected FAD effect and was able to rescue the C. elegans Notch phenotype (Jacobsen et al. 1999). Radiosequencing of the PS1 NTF and CTF identified a major cleavage site between M292 and V293, and the existence of exopeptidase trimming or a possible minor cleavage site between residues M298 and A299 (Podlisny et al. 1997; Steiner et al. 1999b). The mutation M292D resulted in undetectable levels of PS1 NTF in stably-transfected HEK293 cells and also abrogated the FAD effect of the M146L mutation, similar to the effect of mutation at the endoproteolytic site in PS2 (Jacobsen et al. 1999; Steiner et al. 1999b).
PS are present in high molecular weight (HMW) aggregates that range from 100 to 250 kDa, and include both the PS NTF and CTF (Thinakaran et al. 1996; De Strooper et al. 1997; Seeger et al. 1997). FL-PS1 is present in lower molecular weight complexes while the NTF/CTF are present in higher molecular weight complexes in cell culture studies (Capell et al. 1998; Yu et al. 1998, 2000a). However, accumulated endogenous FL-PS1 in the presence of presenilinase inhibitors is capable of forming HMW complexes similar to the NTF/CTF, indicating that the separation seen in overexpression assays may be due to lack of sufficient ‘limiting factor’ required for PS maturation (Beher et al. 2001).
PS1 NTF and CTF can be co-immunoprecipitated and cross-linked, and similarly the PS2 NTF and CTF can be co-immunoprecipitated, indicating that the two PS fragments are physically associated (Capell et al. 1998; Thinakaran et al. 1998). Indeed, the PS1 NTF/CTF can be co-immunoprecipitated from glycerol gradient fractions that contain HMW aggregates. ΔE9 is also able to form HMW complexes, indicating that ΔE9 PS1 may assume a conformation similar to that of an NTF/CTF hybrid PS1, and that only this hybrid PS1 molecule is capable of forming HMW aggregates (Capell et al. 1998).
Expression of NTF alone or CTF alone did not rescue the C. elegans Notch phenotype (Levitan et al. 2001b). NTF constructs (PS1 residues 1–305 or 1–298) were not stabilized in the absence of the PS1 CTF, did not interact with endogenous CTF or replace endogenous NTF expression, and were rapidly degraded by the proteasome, indicating that heterodimer formation may be essential for PS fragment stability (Ratovitski et al. 1997; Citron et al. 1998; Steiner et al. 1998; Saura et al. 1999).
PS1 and PS2 fragments are not co-immunoprecipitated with each other, suggesting that the two molecules form distinct complexes (Saura et al. 1999; Lai et al. 2003). However, a PS1/NTF–PS2/CTF hybrid molecule was subjected to presenilinase cleavage and the two fragments associated to form a stable PS1/NTF–PS2/CTF heterodimer. PS1/NTF–PS2/CTF molecules containing the M146L FAD mutation affected APP metabolism, resulting in an FAD alteration in the ratio of Aβ42/Aβ42+40. Although the two PS do not normally associate, they were capable of interaction in the hybrid molecule after endoproteolysis, suggesting that the PS1/NTF and PS2/CTF formed an intramolecular association prior to endoproteolytic processing (Saura et al. 1999). However, overexpression of separate PS1 NTF and PS1 CTF constructs resulted in the rescue of the C. elegans Notch phenotype. Thus, the NTF and CTF were able to associate despite the lack of a prior association in a FL-PS precursor, supporting the primary importance of the NTF/CTF hybrid molecule for PS function (Levitan et al. 2001b).
The observation that the NTF/CTF heterodimer is the primary species found in vivo, the tight regulation of NTF/CTF production, the rapid turnover of the FL-PS versus the stability of the NTF/CTF, the association of PS NTF/CTF in HMW complexes and the absence of FL-PS in these complexes, and the apparent requirement of PS NTF/CTF heterodimer formation and stabilization for Notch metabolism and the FAD effect on Aβ levels, suggest that FL-PS is an immature precursor and the NTF/CTF PS heterodimer comprises the mature, functional form of the PS protein.
PS1 overexpression leads to a reduction in cytoplasmic β-catenin by facilitating β-catenin degradation and thus, inhibits Wnt signaling through β-catenin (Murayama et al. 1998; Kang et al. 1999; Killick et al. 2001). Glycogen synthase kinase-3β (GSK-3β) also associates with PS1 residues 250–298 (Takashima et al. 1998) and may form a complex with both PS1 and β-catenin (Tesco and Tanzi 2000). However, PS1 regulates β-catenin stability independent of GSK-3β-mediated turnover of β-catenin (Killick et al. 2001). GSK-3β phosphorylates PS1 in the loop region, disrupting PS1 binding to β-catenin. Loss of this interaction has no effect on Aβ production, β-catenin stability or apoptosis, arguing against a functional role for the PS/β–catenin interaction in AD (Kirschenbaum et al. 2001). PS1 also constitutively regulates the phosphorylation of β-catenin on residue p45 by protein kinase A independent of the Wnt signaling pathway (Kang et al. 2002). Loss of this regulation in PS1–/– mice results in the formation of epidermal hyperplasias and carcinomas (Xia et al. 2001). Furthermore, PS1 has been localized to cell–cell contacts, and loss of PS1 leads to a failure to make cell–cell contacts in cell culture (Singh et al. 2001). These data suggest that the interaction between PS1 and β-catenin is not relevant to Aβ production and AD but has a separate function in cellular adhesion.
A role in trafficking of membrane proteins has been proposed for PS. APP trafficking as well as TrkB maturation and autophosphorylation are impaired in PS1–/– neurons, suggesting that TrkB is not correctly trafficked through the secretory pathway (Naruse et al. 1998). Loss of PS in C. elegans causes a reduction in the levels of LIN-12/Notch at the plasma membrane (Levitan and Greenwald 1998). PS may function in facilitating Aβ secretion (Octave et al. 2000). Consistent with that proposal, a PS1 FAD mutation leads to the accumulation of Aβ42 in late compartments of the secretory pathway (Petanceska et al. 2000), and inactivation of PS through mutagenesis causes accumulation of APP at the plasma membrane due to delayed re-internalization (Kaether et al. 2002). A deletion mutation that removes TM1-2 also causes accumulation of APP at the cell surface, but through enhanced surface delivery rather than delayed endocytosis (Leem et al. 2002b). However, lack of PS in PS1–/– fibroblasts does not alter the gross subcellular distribution of APP (Xia et al. 1998; Chen et al. 2000a). It appears that PS functions in trafficking, but the specific proteins it facilitates and the mechanism it utilizes to regulate trafficking are still unclear.
A function that has been investigated extensively is the role that PS plays in the γ-secretase activity that cleaves APP to release Aβ peptides of varying lengths, as well as the CTFγ fragment that may be important for signal transduction and transcription regulation. This review focuses on the proposed function of PS in γ-secretase activity.
Presenilin function in γ-secretase activity
PS Interacts with APP
A role for PS in APP metabolism was suggested from observations that PS FAD mutations cause an increase in the ratio of Aβ42 : Aβ42+40, indicating altered processing at the γ-secretase cleavage site. PS FAD overexpression assays in tissue culture systems confirmed the FAD increase in the ratio of Aβ42 : Aβ42+40 (Xia et al. 1997b). It was subsequently shown that APP co-immunoprecipitates with PS1 and PS2 (Weidemann et al. 1997; Xia et al. 1997a), and that these interactions occur predominantly in the ER but also in the Golgi (Xia et al. 1997a, 1998, 2000). Furthermore, PS1 can co-immunoprecipitate with the APP C99/C83 fragments, the direct substrates of γ-secretase (Verdile et al. 2000; Xia et al. 2000). FAD mutations in PS (PS1/M146L, PS1/C410Y, PS2/N141I, PS2/M239V) or in APP (KM651/652NL, V698F in APP751) do not affect the interaction (Weidemann et al. 1997; Xia et al. 1997a). C-terminal truncations of APP also do not disrupt binding with PS, indicating that the molecules do not interact through the APP cytoplasmic domain (Xia et al. 1997a).
Transgenic mice that express mutant PS transgenes display elevated brain Aβ levels but do not develop neuritic plaques (Duff et al. 1996; Borchelt et al. 1997; Citron et al. 1997; Oyama et al. 1998; Siman et al. 2000). However, mice that overexpress both a PS1 FAD mutant and the APPsw mutant (APP Swedish mutation KM595/596NL) show accelerated Aβ deposition compared with mice expressing the APPsw mutant alone (Borchelt et al. 1997). Plaques in the brains of PS1-FAD/APPsw mice contain numerous Aβ species and are associated with astrogliosis (Holcomb et al. 1998; McGowan et al. 1999). Extensive analyses of Aβ plaques in PS1-FAD/APPsw doubly transgenic mice demonstrate that diffuse and fibrillar Aβ plaques are present in the brains of these animals. The neuritic plaques have dense cores containing primarily Aβ42, and are surrounded by dystrophic neurites and microglia or oligodendrocytes. The mice also have cerebrovascular Aβ deposits (Holcomb et al. 1998; Kurt et al. 2001). Despite the presence of Aβ plaques, these mice do not experience neurodegeneration, and this remains the biggest difference between mouse and human models of AD. Whether or not amyloid deposition causes memory deficits in mouse models of AD remains controversial (Holcomb et al. 1998; Dodart et al. 1999; Chen et al. 2000b; Arendash et al. 2001). These studies in APP/PS1 doubly transgenic mice demonstrate that mutant PS1 enhances Aβ formation and accelerates deposition in the brain, supporting a role for PS1 in the formation of the Aβ peptide.
PS1-deficient mice (PS1–/–) have retarded embryonic growth and die shortly before or immediately after birth (Shen et al. 1997; Wong et al. 1997; De Strooper et al. 1998). Re-introduction of wild-type (wt)-hPS1 or FAD-hPS1 equivalently rescues the PS1–/– mice from embryonic lethality (Davis et al. 1998; Qian et al. 1998). The FAD mutant PS1 causes an increase in Aβ42 levels (Qian et al. 1998), while reduction in the levels of PS1 in PS1+/– heterozygotes does not alter Aβ42 levels compared with wtPS1. Study of brain cultures from PS1–/– animals gave further insight into the role of PS1 in APP metabolism. Cells cultured from day 14 embryos display accumulated APP C99/C83 fragments and a significant reduction in the levels of Aβ or p3 fragments compared with PS1+/+ cells (De Strooper et al. 1998; Xia et al. 1998; Palacino et al. 2000; Wiltfang et al. 2001). Pulse-chase experiments revealed that catabolism of APP C99/C83 fragments is considerably decreased in the PS1–/– cells (Palacino et al. 2000). APPsα and APPsβ fragments are produced and degraded equivalently in PS1+/+ and PS1–/– cells, indicating that α- and β-secretase processing are unchanged by the loss of PS1. Residual Aβ produced in PS1–/– cells could be produced by the endogenous PS2 (De Strooper et al. 1998).
A PS2–/– mouse was created to address the activity of PS2. However, knockout of PS2 had little effect (Donoviel et al. 1999; Herreman et al. 1999). Unlike PS1–/– mice, PS2–/– mice are viable, fertile and have no detectable abnormalities in APP metabolism. The only phenotype identified in the PS2–/– mouse was the development of mild pulmonary fibrosis and hemorrhage as the animals aged (Herreman et al. 1999). The lack of phenotype in the PS2–/– mouse could result from the presence of PS1. To test the functional redundancy of the two PS proteins, a PS1–/–PS2–/– (PS1/2KO) mouse was created. Loss of both PS genes led to embryonic lethality at embryonic day 9.5 (Donoviel et al. 1999; Herreman et al. 1999). Transfection studies in cell lines derived from a PS1/2KO mouse further demonstrated that loss of both PS results in accumulation of APP C99/C83 fragments and the complete absence of Aβ production (Herreman et al. 2000). These results directly implicate PS as the γ-secretase enzyme or an essential co-factor for γ-secretase cleavage of APP C99/C83.
PS interacts genetically with Notch
Another clue to PS function came from the identification of the C. elegans PS sel-12 in a screen for mutants that impact Notch function (Levitan and Greenwald 1995). The SEL-12 protein is expressed throughout the organism, especially in neural cells in the head, pharynx and tail, as well as neurons that control the egg-laying muscles (Levitan et al. 1996; Baumeister et al. 1997). SEL-12 mutations result in a partial loss-of-function egg-laying defective (Egl) phenotype, indicating impairment of Notch function. SEL-12 appears to facilitate signaling of both C. elegans Notch proteins, LIN-12 and GLP-1 (Levitan and Greenwald 1995). Re-introduction of SEL-12 rescues the Egl phenotype of sel-12 mutant worms (Levitan et al. 1996; Baumeister et al. 1997). Expression in sel-12 mutant animals of human PS1 (hPS1) or human PS2 (hPS2) driven off the lin-12 or sel-12 promoters rescued the Egl and vulva defects. PS1 FAD mutations display a partial loss-of-function phenotype with reduced ability to rescue sel-12. Thus, PS are required for Notch function and human PS can substitute for C. elegans PS, indicating conserved function in Notch processing for the human and C. elegans PS (Levitan et al. 1996; Baumeister et al. 1997).
Two homologues of sel-12 were identified in C. elegans based on sequence similarities. Hop-1 (homologue of presenilin) shares 35% identity with sel-12 and 33% identity with hPS1, and is able to rescue the Egl defect of a sel-12 mutant. HOP-1 is not highly expressed and its expression pattern has not been reported (Li and Greenwald 1997). Similar to PS2, hop-1 deletion caused no obvious defects except a decrease in brood size. However, the hop-1;sel-12 double mutant was sterile and demonstrated severe glp-1 and lin-12 loss-of-function phenotypes. Thus, hop-1 and sel-12 may be functionally redundant for Notch signaling in C. elegans (Li and Greenwald 1997; Westlund et al. 1999). Sequence similarities also pinpoint a third putative PS in C. elegans, SPE-4, a protein that functions in spermatogenesis and has a low degree of sequence conservation with SEL-12 (Levitan and Greenwald 1995; Arduengo et al. 1998).
A catalytic cascade activates Notch
The Notch protein functions as a receptor at the cell surface and mediates cell–cell signaling interactions to specify cell fates within an equivalence group, a role that is particularly important during development. Notch is activated by a proteolytic cascade similar to that of APP (Fig 1). During transport through the secretory pathway, Notch is constitutively cleaved by furin in the trans-Golgi network at cleavage site 1 (S1) to form transmembrane-intracellular (TMIC) Notch (Blaumueller et al. 1997; Logeat et al. 1998; Rand et al. 2000). Upon interaction with ligand, furin-cleaved Notch undergoes a second proteolytic cleavage at S2 (site 2). The ectodomain bound to ligand is endocytosed into the ligand-expressing cell (Parks et al. 2000), while the Notch extracellular truncation (NEXT) fragment remains embedded in the membrane (Mumm et al. 2000). NEXT serves as the substrate for a third proteolytic cleavage at S3 (site 3) that releases the Notch intracellular domain (NICD) fragment from the membrane, allowing it to translocate to the nucleus and interact with transcription factors (Jarriault et al. 1995; Schroeter et al. 1998). NEXT is also cleaved at S4 (site 4) in the middle of the transmembrane domain to form the Nβ peptide that is similar to the p3 and Aβ peptides produced from cleavage of APP C83 and C99 (Okochi et al. 2002; Zhang et al. 2002).
S3 cleavage to release NICD is performed by the same presenilin-dependent γ-secretase activity that cleaves APP to release CTFγ. APP and Notch were the first recognized substrates of a novel mechanism to activate receptors for signaling. The first step in regulated intramembranous proteolysis (RIP) (Brown et al. 2000) is cleavage by a metalloprotease (such as TACE for both APP and Notch) that results in shedding of the large ectodomain. The remaining membrane-embedded CTF containing an extracellular domain with less than 30 amino acids is then a substrate for proteolysis within the transmembrane domain by a newly identified class of intramembrane cleaving proteases (I-CLiPs) (Wolfe et al. 1999b) such as γ-secretase. I-CLiP processing releases an intracellular domain (APP CTFγ, Notch NICD) that modulates transcription (Brown et al. 2000).
The PS1/2KO mouse displays a number of severe phenotypes that are similar to the Notch1–/– mouse (Donoviel et al. 1999; Herreman et al. 1999). However, the mice express additional phenotypes that may be indicative of absence of signaling by all four Notch homologues (Notch1–Notch4) (Donoviel et al. 1999) or loss of PS cleavage of other substrates. Transfection studies in cell lines derived from a PS1/2KO mouse confirm that loss of both PS results in the complete absence of NICD production (Herreman et al. 2000). The severity of the PS1/2KO phenotype versus the PS1–/– or PS2–/– single knockouts demonstrates that PS1 and PS2 have overlapping functions and that while PS1 can compensate for loss of PS2, PS2 is unable to compensate for loss of PS1 (Donoviel et al. 1999; Herreman et al. 1999). These data confirm the dependence on PS for γ-secretase cleavage of Notch.
Co-immunoprecipitation studies have further demonstrated that like APP and PS, Notch and PS can interact physically in PS-transfected mammalian cells as well as endogenously in Drosophila (Ray et al. 1999a; Nowotny et al. 2000). PS1 can co-immunoprecipitate FL-Notch and TMIC as well as the γ-secretase substrate ΔE that is similar to the NEXT fragment (Ray et al. 1999a,b). Double staining in mouse brains revealed that PS1 and Notch1 are expressed in the same neuronal populations (Berezovska et al. 1998), and the PS1/Notch interaction can be detected in the ER/Golgi as well as the plasma membrane (Ray et al. 1999b; Ramdya et al. 2003). Interestingly, the PS1–Notch interaction occurs in the presence of γ-secretase inhibitors, indicating that Notch binding to γ-secretase is a separate event from γ-secretase cleavage of Notch (Ramdya et al. 2003). Thus, PS1 and Notch interact early in the secretory pathway and traffic together to the plasma membrane, where Notch signaling occurs through NICD release (Ray et al. 1999b).
The PS1/Notch interaction and the dependence on PS for Notch signaling parallels data on the PS1/APP interaction as well as the defects in APP metabolism observed in PS-deficient cells. γ-Secretase cleavage of APP and S3 release of NICD are both eliminated by the loss of PS (De Strooper et al. 1998). Indeed, inhibitors designed to block γ-secretase activity inhibit NICD production at concentrations that decrease Aβ production and lead to the accumulation of APP C99/C83 fragments (De Strooper et al. 1999). This supports conclusions from APP experiments that PS are directly involved in a γ-secretase activity that cleaves APP and argues that an identical or similar γ-secretase-like activity cleaves Notch.
PS is a putative aspartyl protease
Although in vivo data from C. elegans and transgenic mice strongly support an essential role for PS in γ-secretase activity, little is known about the structure of γ-secretase, the mechanism it utilizes for proteolysis, or the regulation of cleavage. Treatment of cells with difluoroketone and aldehyde peptidomimetic inhibitors designed to mimic the Aβ42 cleavage site cause an accumulation of the APP C99/C83 γ-secretase substrates, increase Aβ42 production at low concentrations, and at high concentrations, block production of Aβ40 and Aβ42 as well as NICD production and nuclear translocation (De Strooper et al. 1998; Wolfe et al. 1998; Berezovska et al. 2000b; Mumm et al. 2000; Beher et al. 2001; Zhang et al. 2001). Further studies with these compounds suggested that the γ-secretase enzyme they block is an aspartyl protease, and hypothesized loose sequence specificity and an α-helical model for the γ-secretase cleavage site in efforts to explain the effects of FAD APP mutations and correlate with the position of the cleavage site within the transmembrane domain (TM) of APP (Wolfe et al. 1999b; De Jonghe et al. 2001).
Additionally, biochemical characterization of γ-secretase in a membrane-based assay using the APP C99 substrate demonstrated that Aβ40 and Aβ42 production have similar pH requirements, and the aspartyl protease inhibitor Pepstatin A inhibited production of both equivalently (Zhang et al. 2001). When inhibitor profiles of seven γ-secretase inhibitors representing a range of structures were compared for APP and Notch in parallel and at numerous concentrations, the rank order of potency was equivalent, implicating a single proteolytic activity in the cleavage of both substrates (Schroeter et al. 2003). The demonstration of competition between APP and Notch for γ-secretase further supports the proposal that a single γ-secretase activity cleaves both substrates (Lleo et al. 2003; Schroeter et al. 2003). Thus, γ-secretase appears to be an aspartyl protease that cleaves both APP and Notch.
Charged residues are rare within TM domains. However, PS have two potentially charged aspartic acid residues in TM6 and TM7 that are completely conserved in all PS (Fig 2). Mutation of either or both of these two residues in PS1 (D257A and D385A) abrogates presenilinase processing within the TM6-TM7 loop. Overexpression of these mutants in CHO cells causes replacement of endogenous PS1 CTFs at levels correlated to the amount of transgene expression (Wolfe et al. 1999a). The levels of holo-APP, APPsα and APPsβ do not change. However, C99/C83 fragments accumulate equivalent to treatment with a peptidomimetic γ-secretase inhibitor and Aβ levels are drastically reduced (Wolfe et al. 1999a; Weidemann et al. 2002). D257A or D385A prevent NICD production and translocation, and fail to rescue the sel-12 phenotype in C. elegans (Brockhaus et al. 1998; Ray et al. 1999b; Berezovska et al. 2000a,b). Similar results were obtained in multiple cell types as well as isolated microsomes. Furthermore, D263A and D366A mutations in PS2 and D375N/A mutations in zebrafish PS all display similar effects on PS endoproteolysis, APP and Notch processing (Leimer et al. 1999; Steiner et al. 1999c; Kimberly et al. 2000). Finally, the D257A mutation in hPS1 is unable to rescue lethality of the PS1–/– mouse, and transfection of D257A into PS1–/– fibroblasts fails to rescue NICD production, APP C99/C83 accumulation or Aβ production in these cells (Nakajima et al. 2000; Xia et al. 2002). Thus, mutation of the conserved aspartates interferes with endogenous wtPS and eliminates both presenilinase and γ-secretase activities.
Similar to stabilized wtPS1 and the ΔE9 mutation that also blocks endoproteolysis, FL-D257A and FL-D385A molecules are found in the high molecular weight complexes (HMW) that contain active wtPS NTF/CTF, as well as the low molecular weight complexes that contain FL-wtPS, suggesting that loss of activity in the aspartic acid mutants is not due to an inability to form the active HMW complex (Yu et al. 2000a; Beher et al. 2001; Tomita et al. 2001; Wang et al. 2004). Aspartic acid mutations do not affect PS1 binding to APP or Notch, and C99/C83 stubs co-immunoprecipitate with PS1 from Golgi fractions, indicating that the lack of activity caused by these mutations is not due to loss of substrate interaction or PS trafficking defects (Ray et al. 1999b; Xia et al. 2000). Finally, aspartic acid mutation is shown to have independent effects on presenilinase and γ-secretase activity, as co-expression of wild-type PS NTF and CTF constructs in C. elegans reconstitutes γ-secretase activity and rescues a sel-12 mutant, while co-expression of an aspartate mutant fragment does not rescue γ-secretase activity (Levitan et al. 2001b). Similar results were obtained in PS1/2KO cells, confirming a direct effect of aspartate mutation on γ-secretase activity (Laudon et al. 2004b). Substitution of an alternative charged residue, glutamate, does not support presenilinase or γ-secretase activity, indicating that aspartate is required in these positions. Additionally, Aβ is generated in microsomes under slightly acidic conditions, consistent with aspartyl protease activity that utilizes two active site aspartates that function optimally at acidic pH. The identification of two essential, completely conserved aspartic acid residues located in adjacent TM domains led to the hypothesis that PS contain the catalytic site for the γ-secretase aspartyl protease activity, and that presenilinase cleavage is an autocatalytic event required for activation of PS (Wolfe et al. 1999a).
The consensus sequence KLGLGDFI surrounding the aspartic acid residue in TM7 of PS is similar to that of the bacterial type-4 prepilin peptidases (TFPP), GMGYGDFK. Thus, PS and TFPP share a consensus G(A)X′GDX′′ (X′ variable, X′′ F >> I > V > L). In particular, the GD is almost completely conserved. Mutation of the aspartic acid residue inhibits TFPP function. Mutation of G384 to K/D in hPS1 inhibits presenilinase endoproteolysis, leads to replacement of endogenous PS2 CTF and the accumulation of APP C99/C83 fragments, and causes a decrease in Aβ and NICD production. There may also be some limited consensus surrounding the aspartic acid residue in TM6, DXXXXLXP (X, variable). TFPP have a similar eight TM structure, although in the opposite orientation to PS. TFPPs remove leader peptides through proteolytic cleavage near the cytoplasmic edge but outside of the membrane, reminiscent of PS cleavage near the cytoplasmic edge within the membrane. Finally, TFPP are aspartyl proteases, and the identification of a conserved active site in PS and TFPP supports the proposal that PS are aspartyl proteases that contain the active site for γ-secretase (Steiner et al. 2000).
PS contain only three regions of high homology, located around each aspartic acid residue and a PAL sequence just C-terminal to TM8 (see Fig 2). Database searches identified five presenilin homologues (PSH), multi-transmembrane proteins that can be aligned with PS in multiple regions. The PSH have a high degree of sequence identity with PS in the three highly conserved regions containing the YD, GhGD and PAL sequences (Fig 4). Like PS, the active site aspartates are predicted to lie within adjacent TM domains, TM6 and TM7. However, unlike PS or TFPP, PSH are predicted to have a 9-TM structure, with an extracellular N-terminus and a cytoplasmic C-terminus. The PAL domain spans the membrane to create the 9-TM PSH molecule that is inserted in the membrane in the opposite orientation from PS (Weihofen et al. 2002; Friedmann et al. 2004). All of the PSH are widely expressed in many tissues, and are present in species ranging from vertebrates, worms and plants to arthropods, fungi and archaea (Grigorenko et al. 2002; Ponting et al. 2002). Thus, PS appear to be members of a larger family of intramembrane-cleaving aspartyl proteases.
PSH3 was identified separately as the signal peptide peptidase (SPP) that is hypothesized to mediate the release of signal peptide fragments from the ER membrane. SPP was identified by labeling with a photo-activated ligand affinity probe (TBL4K) based on an SPP inhibitor. Over 15 homologous proteins were identified in higher eukaryotes and these were classified into five subfamilies of SPP-like proteases (SPPL1-4). The C-termini containing the active site aspartates are conserved in all five subfamilies, while the N-termini are more divergent. SPP contains an ER retrieval signal KKXX and is glycosylated on residues N10 and N20. The TM domains containing the aspartates have opposite orientations in SPP and PS, and SPP substrates are Type II TM proteins that are also inserted in the membrane in the opposite orientation of the PS Type I TM substrates (Weihofen et al. 2002; Friedmann et al. 2004). Similar to γ-secretase substrates, SPP substrates require ectodomain shedding by signal peptidase prior to cleavage by SPP (Lemberg and Martoglio 2002). Despite the similarities between SPP and PS1, in contrast to PS1, SPP does not appear to be activated by endoproteolysis (Martoglio and Golde 2003).
γ-Secretase inhibitors bind to PS
Direct evidence that PS contain the γ-secretase active site, rather than acting as essential co-factors for the γ-secretase protease, was provided by studies using γ-secretase inhibitors designed to bind to and label the active site. The transition state analogue mimic L-685-458 was designed to label the active site of an aspartyl protease (Shearman et al. 2000). It inhibits γ-secretase activity to cleave Notch to form NICD, as well as γ-secretase cleavage of APP C99 to form CTFγ, the APP fragment equivalent to the NICD fragment of Notch (Yu et al. 2001; Weidemann et al. 2002). This demonstrates that CTFγ as well as Aβ and NICD production are PS-dependent, and further designates γ-secretase as an aspartyl protease (Shearman et al. 2000; Yu et al. 2001). Photoactivated forms of this inhibitor (L-852,505 and L-852,646), biotinylated difluoro alcohol transition-state analogues designed specifically to the aspartyl protease active site of γ-secretase, radiolabeled Compound E, and a biotinylated affinity ligand (MerckC) form of the aspartyl protease transition state analogue γ-secretase inhibitor MerckA, all specifically bind PS1 NTF and CTF. The inhibitor also labels the PS2 NTF and CTF, confirming PS2 ability to perform γ-secretase activity. However, PS2 is less sensitive to some of the inhibitors, indicating that subtle differences may exist between the PS1 and PS2 active sites (Esler et al. 2000; Li et al. 2000; Seiffert et al. 2000; Beher et al. 2003; Lai et al. 2003).
FL-PS1 was not labeled, suggesting that FL-PS1 is an inactive zymogen and supporting the hypothesis that the NTF/CTF heterodimer is the functional form of PS (Li et al. 2000; Beher et al. 2003). The FL-PS1 ΔE9 molecule was labeled, consistent with its retention of γ-secretase activity (Li et al. 2000). However, the general aspartyl protease inhibitor pepstatin was able to bind both the FL-PS1 and NTF/CTF, suggesting that the FL-PS1 may have some γ-secretase activity, but the inhibitor had a decreased affinity for ΔE9. PS1 containing an aspartic acid mutation is unable to bind to pepstatin, confirming the requirement of these residues to produce an active catalytic site (Evin et al. 2001). Only a small percentage of total PS1 is associated with functional γ-secretase activity (Beher et al. 2003; Lai et al. 2003), and this activity is located in the Golgi apparatus, compatible with the Golgi localization of the mature NTF/CTF (Beher et al. 2003). Collectively, the results from γ-secretase inhibitor labeling experiments directly identify PS fragments, but not FL-PS, in the formation of the γ-secretase active site, and confirm that aspartic acid mutations result in loss of the functional active site.
It has been proposed that FL-PS may be a zymogen that is activated by autoproteolysis, indicating that the presenilinase activity may be equivalent to the γ-secretase activity (Wolfe et al. 1999a; Li et al. 2000; Yu et al. 2000a). As described above, mutation of the γ-secretase active site aspartic acid residues abolishes presenilinase endoproteolysis of PS as well as γ-secretase activity. Biochemical characterization of the presenilinase indicates that it is an integral membrane protein. Presenilinase functions primarily in the ER and to a limited extent in the Golgi apparatus, consistent with a predominantly ER localization of FL-PS1. Finally, production of PS1 NTF/CTF is efficiently blocked by treatment with the aspartyl protease inhibitor, Pepstatin A, but not inhibitors of other protease classes, is most efficient at slightly acidic pH, and requires an aspartyl protease transition state mimicking moiety for inhibition, indicating that presenilinase is an aspartyl protease (Campbell et al. 2002, 2003). In accord with this, most aspartyl protease transition state analogue inhibitors of γ-secretase also inhibit presenilinase activity, causing an accumulation of the FL-PS precursor (Beher et al. 2001; Campbell et al. 2002, 2003). However, a select group efficiently blocks γ-secretase production of Aβ but has no effect on FL-PS1 accumulation (Beher et al. 2001). Indeed, potent non-transition state analogue inhibitors of γ-secretase have little effect on presenilinase activity (Petit et al. 2001; Campbell et al. 2003). Furthermore, the γ-secretase inhibitor, MW167, is able to inhibit presenilinase more efficiently than it inhibits γ-secretase production of Aβ and CTFγ (Campbell et al. 2003). Thus, presenilinase appears to be an aspartyl protease that may utilize the same active site as γ-secretase, but biochemical distinctions can be made between the two activities.
Some γ-secretase inhibitors that block Aβ production were able to inhibit SPP activity with varying potencies, although DAPT, pepstatin A and JLK2 did not affect SPP. The SPP inhibitor, (Z-LL)2-ketone, did not affect Aβ production. The photoreactive γ-secretase transition state analogue inhibitor, L-852,646, labeled SPP, and other γ-secretase inhibitors displace the binding of TBL4K to SPP, indicating that they are able to compete with TBL4K for binding to SPP (Weihofen et al. 2003). Thus, the active sites of PS and SPP are highly similar, but regulation of proteolysis differs and biochemical distinctions can be made using highly selective inhibitors that distinguish between the two activities (Martoglio and Golde 2003; Weihofen et al. 2003).
Other compounds are able to distinguish between γ-secretase cleavage of APP and γ-secretase cleavage of Notch. The non-peptidic isocoumarin (JLK) γ-secretase inhibitors were reported to inhibit γ-secretase cleavage of APP but not block presenilinase activity or Notch-NICD production (Petit et al. 2001). However, others have reported that these compounds do not directly target γ-secretase (Esler et al. 2002b). Epidemiological studies suggested that chronic use of non-steroidal anti-inflammatory drugs (NSAIDs) may be protective against AD (McGeer et al. 1996; Stewart et al. 1997; in t′Veld et al. 2001; Zandi et al. 2002). Later studies in mice and in cell culture experiments determined that select NSAIDs (including sulindac sulfide, ibuprofen, indomethacin and flurbiprofen) directly target γ-secretase activity to decrease Aβ42 levels and increase Aβ38 levels (Weggen et al. 2001; Morihara et al. 2002; Eriksen et al. 2003; Sagi et al. 2003; Weggen et al. 2003b). Importantly, these compounds do not inhibit APP-CTFγ, Notch-NICD or ErbB4-ICD production by γ-secretase, suggesting that biochemical differences exist between the γ-secretase cleavage that produces Aβ versus that which produces ICD fragments (Weggen et al. 2001; 2003a).
Summary of putative γ-secretase substrates
Numerous substrates have now been identified for the γ-secretase intramembrane-cleaving activity. In addition to APP and Notch, γ-secretase also cleaves epithelial cadherin (E-cadherin), a Type I TM protein that functions in Ca2+-dependent cell–cell adhesion. Proteolysis serves to dissociate E-cadherin–catenin complexes and release β-catenin into the cytoplasm (Marambaud et al. 2002). Another γ-secretase substrate, nectin-1, is a Type I TM protein that is a member of the immunoglobulin superfamily. It mediates Ca2+-independent cell–cell adhesion and also releases β-catenin into the cytoplasm (Kim et al. 2002). Like Notch, the low density lipoprotein receptor-related protein (LRP) is cleaved first by furin, then by a metalloprotease and finally, by γ-secretase to release an ICD that may modulate transcription (May et al. 2002). A fourth Type I TM protein receptor, CD44, also functions in cell–cell adhesion. It is cleaved first by a metalloprotease, leaving a TM-embedded stub similar to APP C99/C83 and Notch NEXT fragments. This stub is subsequently processed by γ-secretase (Lammich et al. 2002; Murakami et al. 2003). γ-Secretase also cleaves ErbB-4, a receptor for epidermal growth factor (EGF). ErbB-4 is a Type I TM receptor tyrosine kinase that functions in cell proliferation and differentiation. In response to binding by TPA or HRG ligands, the metalloprotease TACE cleaves to release a large ectodomain, and the remaining stub is then cleaved by γ-secretase to release an intracellular domain that is translocated to the nucleus (Ni et al. 2001). γ-Secretase similarly cleaves other Type I TM proteins, including the Notch ligands Delta and Jagged (Ikeuchi and Sisodia 2003), and RIP cleavage by γ-secretase may be a general mechanism to activate Type I TM receptors for signaling.
γ-Secretase is a multi-protein complex
Identification of essential complex members
PS presenilinase activity is subject to regulation by a ‘limiting factor’ and the γ-secretase activity appears to reside in a HMW complex, suggesting that other proteins may interact with PS and play an essential role in presenilinase and γ-secretase activities. The first such factor was identified through co-immunoprecipitation studies in HEK cells. Antibodies to PS pulled down Nicastrin (NCT), a single-pass TM protein that is evolutionarily conserved from C. elegans and Drosophila through mammals. NCT was observed in HMW fractions, which included PS1 and PS2, and co-localized with ER and Golgi markers (Yu et al. 2000b; Leem et al. 2002a). Genetic studies in C. elegans showed that NCT is APH-2 (anterior pharynx defective), a protein in the GLP-1/Notch pathway (Goutte et al. 2000; Yu et al. 2000b), and confirmed its interaction with SEL-12/PS (Levitan et al. 2001a). NCT was able to co-immunoprecipitate with APP as well as Notch, and human NCT expression could partially rescue an aph-2 mutation in C. elegans, implying a role for NCT in the metabolism of both γ-secretase substrates (Yu et al. 2000b; Chen et al. 2001; Levitan et al. 2001a).
A screen against a sel-12 partial loss of function mutation in C. elegans that sought to identify PS interactors in the glp-1/Notch pathway pulled out multiple alleles of aph-2/NCT and hop-1. The screen also identified two additional multi-TM, ER/Golgi resident proteins, APH-1 and a novel protein designated PEN-2 (presenilin enhancer). A separate C. elegans screen pulled out aph-1 in addition to several new alleles of glp-1/Notch and aph-2/NCT (Goutte et al. 2002). APH-1 is a seven-TM protein with the C-terminus located in the cytoplasm, while PEN-2 has two TMs with the N- and C-termini located in the lumen/extracellularly (Goutte et al. 2002; Crystal et al. 2003; Fortna et al. 2004). APH-1 may be cleaved within its C-terminus to create fragments of unknown physiological relevance (Fortna et al. 2004). Both APH-1 and PEN-2 interacted genetically with sel-12, hop-1 and aph-2/NCT, and loss-of-function alleles displayed phenotypes similar to those of sel-12 and glp-1 (Francis et al. 2002).
Immunoprecipitation of either APH-1, NCT, PS1 or PEN-2 brought down each of the other three factors, and these immunoprecipitates contained γ-secretase activity that was capable of generating Aβ (Steiner et al. 2002; Baulac et al. 2003; Kimberly et al. 2003; Takasugi et al. 2003). hAPH-1 and hPEN-2 rescued the C. elegans Egl phenotype only when human versions of APH-1, PEN-2 and PS were expressed concomitantly, suggesting sequence divergence between species and stoichiometric interactions between the three proteins (Francis et al. 2002). Indeed, presenilinase and γ-secretase activity were enhanced only when all four factors were overexpressed in mammalian and fly tissue culture assays (Hu and Fortini 2003; Kimberly et al. 2003; Takasugi et al. 2003). Furthermore, presenilinase and γ-secretase activity were reconstituted in yeast (which lacks endogenous PS, NCT, APH-1 and PEN-2, and does not display γ-secretase activity) upon co-expression of all four factors, but no combination of three factors was sufficient to form a functional γ-secretase (Edbauer et al. 2003). Thus, these four proteins are each necessary and together sufficient to perform presenilinase and γ-secretase activities.
Overexpression studies offered further insight into the functions of the complex members (summarized in Figure 5). Overexpression of APH-1 stabilized NCT and was necessary for cell surface localization of NCT (Goutte et al. 2002). It also stabilized PS, leading to increased levels of FL-PS1 and PS1 fragments, and this effect was augmented by co-expression of NCT, although NCT alone had no effect on FL-PS1 levels (Hu and Fortini 2003; Takasugi et al. 2003). Mutation analysis implicated a GXXXG helix interaction domain in the fourth TM domain of APH-1 for its interaction with PS, NCT and PEN-2 (Lee et al. 2004). Endogenous and overexpressed APH-1 is able to interact with iNCT and mNCT, and both FL-PS1 and PS fragments in HMW complexes that contain γ-secretase activity (Gu et al. 2003). It has been proposed that APH-1 may be a scaffold for the PS/NCT interaction (Hu and Fortini 2003). In support of this, overexpression of APH-1 and NCT caused an accumulation of APH-1, but the addition of PS resulted in a decrease in APH-1 levels (Hu and Fortini 2003). Furthermore, a minor subcomplex containing mNCT, APH-1 and PS1 CTF was identified (Fraering et al. 2004), a stable subcomplex of iNCT and APH-1 independent of PS and PEN-2 has been observed (LaVoie et al. 2003; Shirotani et al. 2004), and APH-1 was only isolated with the γ-secretase complex under weak detergent conditions, indicating that it might dissociate from the complex upon maturation (Culvenor et al. 2004).
Expression of PEN-2 caused an increase in levels of PS fragments (Luo et al. 2003; Takasugi et al. 2003), and PEN-2 bound preferentially to FL-PS1 and mNCT (LaVoie et al. 2003), suggesting a role for PEN-2 in complex maturation. PEN-2 and PS1 NTF were isolated as a minor subcomplex within the active γ-secretase complex (Fraering et al. 2004), and deletion, domain swap and mutagenesis studies have implicated several PEN-2 domains in the PEN-2/PS interaction (Crystal et al. 2003; Hasegawa et al. 2004; Kim and Sisodia 2004). Unlike APH-1 overexpression, co-expression of PEN-2 with PS, NCT or APH-1 alone had no effect on PS, NCT or APH-1 stability. However, addition of PEN-2 with APH-1/PS/NCT resulted in the production of PS fragments (Hu and Fortini 2003).
Together, the data from knockdown and overexpression studies suggest that APH-1 and NCT form an initial complex scaffold to which FL-PS binds, resulting in stabilization of the FL-PS molecule. PEN-2 is then added to the complex and promotes complex trafficking and maturation (PS presenilinase endoproteolysis, NCT glycosylation) and γ-secretase activity. However, recent data show that complexes containing ΔE9 require PEN-2 for maturation, indicating that it is generally required for complex maturation, independent of a possible role in regulation of presenilinase cleavage, and may be required for the stabilization of PS in its active conformation (Prokop et al. 2004). Continuing research will further elucidate the regulation of γ-secretase complex formation and the function that each molecule performs in the formation, maturation and activity of the complex.
Structural requirements for presenilin function
PS dimerization in γ-secretase activity
The presenilin homologue SPP is an approximately 42 kDa protein. However, predominantly 95 kDa protein bands are recognized through sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS–PAGE) and western blotting for SPP. Heat or acid treatment of cell lysates reduces this protein complex to a 45 kDa band, suggesting that the 95 kDa band might be a dimerized form of SPP. This was confirmed by co-purification and subcellular co-localization of two differentially tagged SPP molecules (V5his or FLAG tag) co-expressed in cell culture. The homodimer forms rapidly following translation and a portion of SPP homodimer remains stable up to 24 h, whereas the SPP monomer has a half-life of 2 h and none is visible at 24 h. Additionally, an active site-directed photoaffinity γ-secretase inhibitor bound specifically to the dimeric, but not to the monomeric, form of SPP. These data suggest that SPP exists primarily as a homodimer in vivo and that the dimeric form of SPP is the mature and functional form of the protein (Nyborg et al. 2004).
SPP cleavage can be reconstituted in yeast without the addition of co-factors (Weihofen et al. 2002). However, γ-secretase requires four proteins, APH-1, NCT, PS and PEN-2, that associate in the γ-secretase complex in an unknown stoichiometry. Like SPP, PS may exist as a multimer in the active γ-secretase complex. The simplest complex would contain a PS dimer in addition to monomers of APH-1, NCT and PEN-2. Yeast two-hybrid experiments, yeast split-ubiquitin assays, mammalian co-immunoprecipitation studies and cross-linking experiments with a photoreactive γ-secretase inhibitor in mammalian cells demonstrate that PS1 can form FL/FL, NTF/NTF and CTF/CTF homodimers as well as the NTF/CTF heterodimer (Cervantes et al. 2001; Hebert et al. 2003a,b; Schroeter et al. 2003). FL-PS1 containing a FAD or aspartic acid mutation retains the ability to form a FL/FL homodimer, indicating that dimerization may precede activation within the γ-secretase complex (Hebert et al. 2003a,b). The functional interaction within the homodimer could be in active site formation or, alternatively, one monomer could bind substrate and present it to the active site of a second monomer for γ-secretase cleavage (Schroeter et al. 2003). Thus, the active sites of both SPP and PS may be formed by homodimers of each protein, and dimerization may be important for γ-secretase activity.
Functional domains in PS
Identification of PS domains that are required for its processing, trafficking, protein interactions, presenilinase and γ-secretase activities will aid understanding of the mechanism of FAD mutations and may indicate possible methods to counteract their effects. The distribution of the PS1 FAD mutations does not highlight a particular domain important for PS1 activity because the FAD mutations are spread throughout the molecule. However, they are located primarily in the eight TM domains. Deletion and mutation studies have begun to give some insight into the PS domains that are required for its proper maturation and function.
Studies of PS1 dimerization suggest domains that are important for PS/PS associations. Yeast two-hybrid and GST pull-down experiments indicate that homodimer and heterodimer interactions are mediated by the cytoplasmic N-terminus, the hydrophilic loop between TM1 and TM2 (HL1) in conjunction with TM2, the large cytoplasmic loop between TM6 and TM7, and the final 61 C-terminal residues (Hebert et al. 2003a). FAD mutations in an HL1 peptide impair homodimerization, suggesting that FAD mutations could cause subtle changes in binding affinities that might contribute to PS FAD phenotypes (Cervantes et al. 2001). Substitution of HL5 for HL1 enhances the NTF/NTF interaction but eliminates presenilinase and γ-secretase activity while stabilizing the FL-PS, indicating that HL1 may also function in promoting PS activity (Cervantes et al. 2004).
Murphy et al. (2000) and Leem et al. (2002b) both characterized a PS1 mutant in which TM1, TM2 and their connecting loop were deleted (ΔM1-2). This deletion resulted in inefficient presenilinase endoproteolysis and stabilization of the FL-ΔM1-2 molecule, decreased γ-secretase activity, abnormal APP metabolism, enhanced trafficking of APP to the cell surface, and decreased NCT maturation and trafficking to the cell surface. Neither study narrowed in on the specific domain(s) within this large region that mediated these effects. Annaert et al. (2001) identified an interaction between PS1 and telencephalin (TLN) using a yeast two-hybrid assay and GST pull-down experiments. TLN facilitates long-term potentiation and also functions in neurite outgrowth (Tian et al. 2000; Nakamura et al. 2001). PS interaction with TLN required TM1, TM2 and the extreme cytoplasmic C-terminus. These same regions were able to interact with APP (Annaert et al. 2001). It appears that the N-terminus of PS may contain multiple protein interaction domains that mediate associations between two PS molecules, PS and γ-secretase complex members, and PS and substrates.
The large cytoplasmic loop between TM6 and TM7 has been used for numerous yeast two-hybrid experiments. However, most of these interactions were not replicated and no new data have surfaced since the original reports. Deletion of the loop (with retention of the endoproteolytic site) has no effect on endoproteolysis or PS fragment dimerization, does not affect Aβ production from APP and does not impair NICD production from Notch. Thus, the loop does not have an essential role in PS/γ-secretase function (Saura et al. 2000). However, mutations of the Y288 residue N-terminal to the endoproteolytic site have varying effects on PS presenilinase and γ-secretase activities, although all mutants form γ-secretase complexes (Laudon et al. 2004a).
Differing effects on presenilinase endoproteolysis, APP cleavage and Notch cleavage have been identified for very few PS mutations, as most mutations affect all PS activities equivalently. In addition to Y288, mutations at L286, just two residues N-terminal to Y288, also separate γ-secretase function through differential effects on Aβ production and Notch cleavage (Kulic et al. 2000). The L271V mutation results in increased expression of PS1 lacking exon eight (ΔE8), an inactive molecule that can no longer bind to tau or GSK-3β, indicating that these proteins interact with PS1 through an exon8/exon9 domain (Kwok et al. 2003). Mutations at residue L166 in TM3 also have varying effects on presenilinase versus γ-secretase activity, as well as APP-Aβ versus APP-CTFγ and Notch S3 production (Moehlmann et al. 2002). Finally, TM5 mutations in both PS1 and PS2 are also capable of distinguishing γ-secretase cleavage of APP versus Notch. The difference in γ-secretase function caused by these mutations is not due to differences in substrate binding, suggesting that it is likely due to conformational changes that affect γ-secretase cleavage of APP but not Notch (Walker et al. 2005).
Mutational analysis has identified a few domains that distinguish between PS activity in presenilinase, APP and Notch cleavage. However, the mechanism behind these differences is, in most cases, unknown. Indeed, details of the structural requirements for PS functions, and the mechanisms of PS presenilinase and γ-secretase activities, are just beginning to be understood. The most important questions that remain pertain to PS structure, as it is likely that subtle alterations in PS conformation cause the FAD phenotype and are also responsible for the PS-dependent differences observed in presenilin, APP and Notch metabolism. Understanding the mechanism and regulation of the presenilinase and γ-secretase activities may allow the development of therapeutics that can modify the risk for AD by altering the metabolism of APP.